Query answering with controlled access knowledge graph

Patent metadata

Field	Value
Publication number	US-11062700-B1
Application number	US-201816218836-A
Country	US
Kind code	B1
Filing date	Dec 13, 2018
Priority date	Dec 13, 2018
Publication date	Jul 13, 2021
Grant date	Jul 13, 2021

Abstract

A method comprising receiving first data representative of a query received by an electronic device. Device characteristic data indicative of at least one characteristic of the electronic device is received. It is determined, using the device characteristic data, that the electronic device is authorized to access a first portion of at least one knowledge graph, which is an access-controlled portion. The at least one knowledge graph also includes a second portion which is a non-access-controlled portion. The first data is sent to at least the first portion of the at least one knowledge graph. Second data is received from the first portion of the at least one knowledge graph. The second data is representative of an answer to the query. Answer data representative of the answer to the query is generated using the second data.

First claim

What is claimed is:

1. A method comprising:
receiving first data representative of a first query received by a first electronic device;
receiving first device characteristic data indicative of at least one characteristic of the first electronic device;
determining, using the first device characteristic data, that the first electronic device is authorized to access a first portion of at least one knowledge graph, wherein the first portion of the at least one knowledge graph is an access-controlled portion and the at least one knowledge graph comprises a second portion which is a non-access-controlled portion;
sending the first data to at least the first portion of the at least one knowledge graph;
receiving second data from the first portion of the at least one knowledge graph, wherein the second data is representative of an answer to the first query;
generating first answer data representative of the answer to the first query, using the second data;
receiving third data representative of a second query received by a second electronic device;
receiving second device characteristic data indicative of at least one characteristic of the second electronic device;
determining, using the second device characteristic data, that the second electronic device lacks authorization to access the first portion of the at least one knowledge graph;
sending the third data to the second portion of the at least one knowledge graph;
receiving fourth data from the second portion of the at least one knowledge graph, the fourth data representative of an answer to the second query; and
generating second answer data representative of the answer to the second query, using the fourth data.

2. The method according to claim 1 , wherein the answer to the first query is a first answer to the first query, and the method comprises:
receiving fifth data representative of a second answer to the first query, different from the first answer to the first query, from the second portion of the at least one knowledge graph; and
prioritizing the first answer to the first query over the second answer to the first query on the basis that the first answer to the first query is received from the first portion of the at least one knowledge graph.

3. The method according to claim 1 , wherein the first device characteristic data comprises at least one of: device identification data indicative of an identity of the first electronic device, user profile data indicative of a user profile associated with the first electronic device, organization identification data indicative of an organization associated with the first electronic device, context data indicative of a context in which the first query was received, device configuration data indicative of at least one configuration aspect of the first electronic device, or application identification data indicative of at least one application operative on the first electronic device.

4. The method according to claim 1 , wherein the first electronic device is an audio capture device, the first data is first audio data representative of a spoken utterance captured by the audio capture device, the device characteristic data comprises user profile data indicative of a user profile associated with the audio capture device, and the method comprises:
generating the first device characteristic data by processing the first audio data using voice recognition to identify the user profile associated with the audio capture device;
generating confidence data representative of a confidence with which the user profile associated with the audio capture device is identified; and
determining to send the first data to at least the first portion of the at least one knowledge graph using the confidence data.

5. The method according to claim 1 , wherein the first at least one portion of the at least one knowledge graph is a first knowledge graph associated with a first plurality of user profiles and the second at least one portion of the at least one knowledge graph is a second knowledge graph associated with a second plurality of user profiles comprising the first plurality of user profiles and at least one further user profile.

6. The method according to claim 1 , comprising:
receiving sixth data representative of a third query received by the first electronic device;
sending the sixth data to the at least one knowledge graph;
determining that the first portion of the at least one knowledge graph lacks an answer to the third query;
receiving seventh data from the second portion of the at least one knowledge graph, wherein the seventh data is representative of a fourth answer to the third query;
generating third answer data representative of the fourth answer to the third query, using the seventh data.

7. The method according to claim 1 , wherein the first portion of the at least one knowledge graph is a first knowledge graph, the second portion of the at least one knowledge graph is a second knowledge graph, and the method comprises sending the first data to the second knowledge graph before the receiving the second data from the first knowledge graph.

8. The method according to claim 1 , wherein the first portion of the at least one knowledge graph is a first knowledge graph, the second portion of the at least one knowledge graph is a second knowledge graph, and the first knowledge graph is authorized to access the second knowledge graph but the second knowledge graph is prevented from accessing the first knowledge graph.

9. The method according to claim 1 , comprising:
determining that the answer to the first query involves an attribute which is absent from the first portion of the at least one knowledge graph;
identifying, using the second portion of the at least one knowledge graph, the attribute associated with an entity for which the attribute is to be obtained;
obtaining, from the second portion of the at least one knowledge graph, attribute data representative of the attribute; and
using the attribute data to generate the second data.

10. The method according to claim 1 , comprising:
receiving entity text data from the second portion of the at least one knowledge graph, the entity text data representative of entity text associated with an entity of the first query;
using the entity text data to identify that the first query comprises the entity.

11. The method according to claim 1 , comprising:
receiving query template data representative of a template of an exemplar query answerable by the first portion of the at least one knowledge graph and the second portion of the at least one knowledge graph;
identifying that a structure of the first query corresponds with a structure of the exemplar query; and
querying the first portion of the at least one knowledge graph using the template of the exemplar query to generate the second data.

12. The method according to claim 1 , comprising:
receiving relationship data from the second portion of the at least one knowledge graph, the relationship data indicative of a relationship between a first entity of the first portion of the at least one knowledge graph and a second entity of the first portion of the at least one knowledge graph; and
associating, within the first portion of the at least one knowledge graph, the first entity with the second entity via the relationship.

13. The method according to claim 1 , comprising:
generating the first portion of the at least one knowledge graph in accordance with an ontology template representative of a pre-defined relationship between a first entity class and a second entity class;
receiving structured data;
determining that the structured data comp

Assignees

• Amazon Tech Inc

Inventors

• Azimi Pooria
• Pearson Feinn Zachary Philip
• Reno Thomas
• Segal Ivgeni
• Zucchi John

Classifications

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N20/10

  using kernel methods, e.g. support vector machines [SVM] · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title

• G06N3/08

  Learning methods · CPC title